Stemming from Wand and Weber’s (2002) comments on the need to study the impact of representational deficiencies on the effectiveness, usefulness and/or efficiency of a modelling language, we have sought to study the impact of ontological completeness and clarity on the perceived usefulness and ease of use of a language. As such, we have restricted our investigation in the sense that we do not consider other related phenomena such as, for instance, the quality of the model produced. We acknowledge that other areas of evaluation remain in which the consequences of representational deficiencies still need to be explored.
In this context of acceptance, the technology acceptance model (TAM) (Davis, 1986, 1989) postulates, and it has been shown in an extensive number of empirical studies, that perceived usefulness (PU) and perceived ease of use (PEOU) of an IS artifact directly influence an individual’s intention to use that IS artifact (Davis, 1989; Davis et al., 1989; Moore and Benbasat, 1991). Such intention in turn has been found to accurately predict the actual use of the artifact (Davis et al., 1989; Venkatesh and Davis, 1996).
Hence, we see an opportunity to converge, if not amalgamate, two of the most influential approaches to IS research. The extensive amount of research related to TAM has reportedly made it one of the most influential and commonly employed IS models (Lee et al., 2003; King and He, 2006). Its advantages include the parsimony and explanatory power of the model (Venkatesh and Davis, 2000) and the well-researched and validated measurement inventory with high levels of reliability and validity of constructs and measurement scales (Davis, 1989; Segars and Grover, 1993). The large number of TAM studies will not be recapitulated here; instead the reader is referred to an annotated overview such as that given in, for instance, Lee et al., (2003). One interesting point, however, must be made. King and He (2006) found in their rigorous meta-analysis of TAM that, despite its recent adaptations to, for example, the method context (Moody, 2003), extensions such as the TAM2 model (Venkatesh and Davis, 2000), and revisions such as the UTAUT model (Venkatesh et al., 2003), the original model nevertheless is of high reliability, has good explanatory power and obtains high levels of robustness. We therefore deem TAM, in its original form, a suitable starting point for our line of investigation.
The interesting observation to be made with respect to representation theory is that TAM specifies a general model of IS acceptance that needs to be tailored to the specific research context (Fichman, 1992). As we, in our research, are concerned with conceptual modelling and the languages used for such efforts, we see an opportunity to link these two theories to study the acceptance of modelling languages. Along similar lines, Venkatesh and Davis (1996, 2000) argue that it is necessary to better understand the determinants of PU and PEOU since the generality of TAM, which allows for wide applicability, induces a lack of focus on the particular artifact under observation. Accordingly, we explicitly explore the determinants of PU and PEOU in the context of conceptual modelling languages by drawing on the principles of representational analysis.
Starting with PU, Moody (2003) argues that the original definition of PU (Davis, 1989) must be extended to reflect the objectives of the particular task for which the artifact is being used. Adopting this insight in the context of conceptual modelling, we can perceive PU as ‘the degree to which a person believes that a particular language will be effective in achieving the intended modelling objective’. This definition reflects the notion of rational selection (Rescher, 1973), which states that, generally, those methods will be adopted that outperform others or are more effective in achieving intended objectives. Based on this understanding, we can argue that ‘good’ languages are those that contain all the constructs needed to produce complete representations of the relevant phenomena in a real-world domain of interest (Weber, 1997). Clearly, the notion of a complete language (without construct deficit) reflects the notion of an effective language with respect to the objective of conceptual modelling to build a representation of selected phenomena in the problem domain (Mylopoulos, 1992; Wand and Weber, 2002; Siau, 2004). Accordingly, we urge that ontological completeness is a determinant of the PU of a conceptual modelling language (see Figure 3), based on the argument that PU represents a perceptual judgment of an artifact’s effectiveness (Rescher, 1973).
PEOU, adapting its original definition in Davis (1989) to the context of conceptual modelling, can be understood as ‘the degree to which a person believes that using a particular language will be free of effort’. Modelling ‘free of effort’ means modelling without complexity (Gemino and Wand, 2005), which in turn provides another link to representation theory. Weber (1997) argues that, in addition to the question of ‘what’ can be represented, also the question of ‘how’ it can be represented is of importance. He says that the clarity of a language is determined by how unambiguously the meaning of its constructs is specified and thus how much effort is needed to apply desired real-world meaning to them. The notion of clarity embraces the three situations of construct overload, redundancy and excess. That is, a formative relationship exists between these sub-constructs and the overall construct of ontological clarity. Again, one can perceive a link between the notion of clarity of a language and PEOU of a language with respect to the aim of conceptual modelling to facilitate communication and understanding among stakeholders (Mylopoulos, 1992; Wand and Weber, 2002; Siau, 2004). Consequently, we argue that ontological clarity is a determinant of PEOU of a language (see Figure 3).
Aside from these primary constructs of the research model, in every scientific study it is necessary to identify and take into account endogenous variables that potentially impose a strong contingent effect on the ‘independent variable — dependent variable’ relationship. Moderating variables must be identified based on the context (Fichman, 1992). We draw on variables that have previously been identified, and validated, as having consequences for our particular research context. Previous representational analyses of process modelling languages (see above) have identified and explored the contextual factors of modelling role, modelling purpose, modelling tool, modelling conventions and modelling experience, all of which can moderate the perceived criticality of representational deficiencies, and which we therefore include in our model (see Figure 3).
Aside from these contextual factors, we also draw on one of the most frequently noted limitations of previous TAM studies, namely the impact of ‘voluntariness’ on adoption decisions. Moore and Benbasat (1991) first recognised that the acceptance behaviour of individuals may be influenced by a mandate from superiors, which is expressed as a moderating effect of a variable ‘voluntariness’. This has been included in some studies (e.g. Venkatesh and Davis, 2000; Venkatesh et al., 2003). In the case of conceptual modelling, we note that in most cases the use of a particular modelling language is indeed mandated in organisations by superiors such as modelling coaches, consultants or other influential individuals. Accordingly, we argue that the extent of voluntariness impacts the causal relationship between the intention to use a modelling language and the actual usage of the language.
Figure 3 shows the overall research model, adapted to our selected research case of the BPMN modelling language. In previous work (Recker et al., 2006) we have identified and empirically tested representational deficiencies of BPMN with respect to construct deficit, redundancy, overload, and excess, and these results will now be used to derive measurement items for each representational deficiency.
After the formulation of the research model we need to operationalise the hypotheses and measurement items contained in the model to create an empirical instrument with which to test it. The level of dissemination and maturity of TAM, and its measurement inventory, allows us to develop an appropriate instrument by adopting existing measurement scales to the context of process modelling languages. Nevertheless, this task still poses a number of challenges. Most significantly, several researchers have noted limitations related to the conceptualisation of ‘usage’ (DeLone and McLean, 2003) and the use of self-reported measurements (Lee et al., 2003). Also, the definition of ‘intention to use’ must be slightly modified to ‘intention to continue to use’. This adaptation reflects the fact that only when a modelling individual has started using a language for modelling tasks is he or she able to explore its potential representational deficiencies and form an opinion about its usefulness and ease of use. Second, we will convert the measurement instrument to a web-based survey and distribute it to both actual and potential adopters of BPMN. In order to account for the fact that user perceptions and intentions may change over time (Lee et al., 2003) we will also add a longitudinal aspect to the study by measuring these quantities at two points in time: (a) in a period of early adoption and exposure to BPMN, and (b) in a later period of increased familiarity with the modelling language. This should allow us not only to counter the criticism of most acceptance studies that they are restricted to cross-sectional studies (Agarwal and Karahanna, 2000), but also account for, and further explore, the moderating effect of modelling experience on representational deficiencies and their impact on language acceptance. Also, it should allow us to study the impact of representational deficiencies not only on an individual’s early intention to start to use a modelling language but also on the decision to continue to use it after a period of prolonged exposure. Finally, a web-based format of the instrument permits the gathering of data from a multitude of potential respondents across different regions and cultures, thereby overcoming the bias of restricted contextual settings and supporting potential cross-contextual analyses as well.
We would like to note here an obvious limitation of this proposed research. The presented study draws heavily on the principles of representation theory and TAM. Hence, the focus of study is restricted by the filtering lenses that these models employ. Accordingly, the research model may lack other, potentially relevant, endogenous variables that may also affect user acceptance of modelling languages. Nevertheless, the scope of the proposed model enables us to focus work on gaining insights into the expressiveness of the combination of the two theories, and to thereby avoid the necessity to translate findings from different theoretical bases.